QACNN: Query-based Attention CNN
Implementation of my research:
Query-based Attention CNN for Text Similarity Map (https://arxiv.org/abs/1709.05036)
6 folders and 1 main script should be included :
-
preprocess
- plot2vec.py
- qa2vec.py
-
word_vec
- register.json
- glove.42B.300d.json
-
raw_data
- plot
- IMDB_KEY.split.wiki
- question
- qa.json
- plot
-
output_data
- plot
- question
-
utility
- utility.py
-
model
- MODEL.py
-
main.py
usage
- put the data given by MovieQA into raw_data folder (including IMDB_KEY.split.wiki and qa.json)
- download GloVe word2vec and put it into word_vec folder (GloVe: https://drive.google.com/file/d/0B3UsyrHYzsTzekMtbndMZXB1Ync/view?usp=sharing)
- run 2 scripts (plot2vec.py & qa2vec.py), and all the preprocessed data will automatically be saved to output_data folder
- run main.py